Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> writes:
> On Thu, 2 Jun 2016 15:09:49 +0530 "Aneesh Kumar K.V" <aneesh.kumar@xxxxxxxxxxxxxxxxxx> wrote:
>
>> Now that we track page size in mmu_gather, we can use address based
>> tlbie format when doing a tlb_flush(). We don't do this if we are
>> invalidating the full address space.
>>
>> ...
>>
>> void radix__tlb_flush(struct mmu_gather *tlb)
>> {
>> + int psize = 0;
>> struct mm_struct *mm = tlb->mm;
>> - radix__flush_tlb_mm(mm);
>> + int page_size = tlb->page_size;
>> +
>> + psize = radix_get_mmu_psize(page_size);
>> + if (psize == -1)
>> + /* unknown page size */
>> + goto flush_mm;
>> +
>> + if (!tlb->fullmm && !tlb->need_flush_all)
>> + radix__flush_tlb_range_psize(mm, tlb->start, tlb->end, psize);
>> + else
>> +flush_mm:
>> + radix__flush_tlb_mm(mm);
>
> That's kinda ugly. What about
>
> void radix__tlb_flush(struct mmu_gather *tlb)
> {
> int psize = 0;
> struct mm_struct *mm = tlb->mm;
> int page_size = tlb->page_size;
>
> psize = radix_get_mmu_psize(page_size);
>
> if (psize != -1 && !tlb->fullmm && !tlb->need_flush_all)
> radix__flush_tlb_range_psize(mm, tlb->start, tlb->end, psize);
> else
> radix__flush_tlb_mm(mm);
> }
>
> ?
>
> We lost the comment, but that can be neatly addressed by documenting
> radix_get_mmu_psize() (of course!). Please send along a comment to do
> this and I'll add it in.
I will update the patch. But this patch (Patch 4) need to go through
powerpc tree because radix__flush_tlb_range_psize is not yet upstream.
As I mentioned in the previous thread, if you can take patch 1 to patch 3 that
will enable wider testing w.r.t other archs and ppc64 related changes can
go later via powerpc tree ?
>
> --- a/arch/powerpc/mm/tlb-radix.c~powerpc-mm-radix-implement-tlb-mmu-gather-flush-efficiently-fix
> +++ a/arch/powerpc/mm/tlb-radix.c
> @@ -265,13 +265,9 @@ void radix__tlb_flush(struct mmu_gather
> int page_size = tlb->page_size;
>
> psize = radix_get_mmu_psize(page_size);
> - if (psize == -1)
> - /* unknown page size */
> - goto flush_mm;
>
> - if (!tlb->fullmm && !tlb->need_flush_all)
> + if (psize != -1 && !tlb->fullmm && !tlb->need_flush_all)
> radix__flush_tlb_range_psize(mm, tlb->start, tlb->end, psize);
> else
> -flush_mm:
> radix__flush_tlb_mm(mm);
> }
> _
>
> I'll await feedback from the other PPC developers before doing anything
> further on this patchset.
>
> hm, no ppc mailing lists were cc'ed. Regrettable.
I missed that. I can resend the series again adding ppc-devel to cc: ?
-aneesh
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>